Um método de seleção de causas básicas de óbito por meio das distribuições de diagnósticos mencionados
Data
2024-03-26
Autores
Tardelli, Adalberto Otranto 
Orientadores
Paiva, Paulo Bandiera
Tipo
Tese de doutorado
Título da Revista
ISSN da Revista
Título de Volume
Resumo
Introdução: As estatísticas de mortalidade são produzidas na maioria dos países, segundo as regras definidas pela OMS nas sucessivas revisões da Classificação Estatística Internacional de Doenças e Problemas Relacionados à Saúde, atualmente em sua 10ª Revisão (CID-10). Têm como eixo principal a chamada Causa Básica do óbito, selecionadas de acordo com as Regras Internacionais de Seleção de Causas Básicas, cuja interpretação demanda conhecimento de relações causais entre doenças e uma considerável capacidade técnica para aplicá-las correta e sistematicamente.
Objetivo: O estudo propôs identificar os conjuntos de diagnósticos informados nos dados das Declarações de Óbito processadas pelo Sistema de Informações de Mortalidade (SIM) brasileiro, associá-los com as correspondentes causas básicas selecionadas e avaliar o grau de acurácia de um método estatístico de previsão destas.
Métodos: Os dados das Declarações de Óbito (DO) do SIM com os dados do Estado de São Paulo ocorridos no período de 2000 a 2018 foram carregados num Repositório de dados de acesso aberto, especificado e desenvolvido com a capacidade de recuperação de dados requerida pelo estudo. Foram selecionados os casos com os códigos de causa básica de óbito (CB) e de cada diagnóstico mencionado (DM) considerados válidos segundo a lista de códigos CID-10 distribuída na página de transferência de arquivos do SIM e com um mínimo de 2 e um máximo de 6 DMs. Foram identificados todos os conjuntos de DMs dos dados relativos aos períodos 2009-2013 e 2014-2018, representando-os, para cada DO, em seis tipos de conjuntos de DMs, com os quais criaram-se regras quantitativas associando cada conjunto de DMs às correspondentes CBs (Regras MD=>UCD) a partir das respectivas distribuições proporcionais. As Regras do período 2009-2013 foram, então, aplicadas a cada DO dos óbitos ocorridos no período 2014 a 2018. E vice-versa. Por fim, as causas básicas previstas pelo método foram comparadas às selecionadas pelo SIM.
Resultados: Nos dados de 2009-2013, foram selecionados 5218 distintos códigos de causas básicas. Foram identificados os seguintes conjuntos de diagnósticos mencionados: i) 635.995 conjuntos distintos de DMs considerando as linhas do chamado atestado de óbito e a ordem onde foram mencionados; ii) 596.544 desconsiderando a posição, iii) 535.212 desconsiderando também a ordem das menções; iv) 336.119 combinações de pares de DMs; v) 270.698 pares distintos quando desconsiderada a ordem das menções; e vi) 7.430 distintos códigos de diagnósticos. Combinados com as respectivas CBs, ocorreram 4.315.054 distintos conjuntos de DMs x CBs, sendo 637.196, 611.208, 558.017, 1.196.629, 1.081.936 e 230.068 correspondentes a esses seis tipos de conjuntos de DMs. Ao aplicar as Regras MD=>UCD do período 2009-2013 aos dados de 2014-2018, foram atribuídas causas básicas a 99,99% dos casos com 89.99% de acerto, ao nível de Subcategoria CID-10. Com as Regras MD=>UCD do período 2014-2018 aplicadas aos dados de 2009-2013, foram atribuídas causas básicas a 99,99% dos casos com 90.36% de acerto, igualmente ao nível de Subcategoria CID-10.
Conclusão: Foi disponibilizado um Repositório de dados das Declarações de Óbito registradas no Brasil ao longo dos anos 2000 a 2018, em acesso aberto e com capacidade de recuperação de dados adequada também a estudos de causas múltiplas de morte. O estudo concluiu que a causa básica de morte de uma DO está fortemente associada não apenas com a informação contida na representação estruturada dos diagnósticos mencionados, mas também com outras representações destes, tais como sequências e pares de DM, ainda que desconsiderando a posição e/ou a ordem em que foram declarados na DO: causas básicas de óbito foram previstas com considerável grau de acurácia, simplesmente por meio das distribuições proporcionais das sequências, pares e unidades de diagnósticos mencionados e suas correspondentes causas básicas. As causas básicas “previstas” poderão servir de guia para verificação manual das causas básicas selecionadas nos sistemas de informação de mortalidade e podem ser úteis para revelar casos que exigem especial atenção – inclusive nas tabelas de decisão dos sistemas automatizados de seleção de causas de morte.
Introduction: Mortality statistics are produced in most countries, according to the rules defined by the WHO in successive revisions of the International Statistical Classification of Diseases and Related Health Problems, currently in its 10th Revision (ICD-10). Their main axis is the so-called Underlying Cause of death, selected in accordance with the International Rules for Selection of Underlying Causes, whose interpretation requires knowledge of causal relationships between diseases and considerable technical capacity to apply them correctly and systematically. Objective: The study proposed to identify the sets of diagnoses reported in data from Death Certificates processed by the Brazilian Mortality Information System (SIM), associate them with the corresponding selected underlying causes and evaluate the degree of accuracy of a statistical prediction method of these. Methods: Data from SIM Death Certificates (DO) with data from the State of São Paulo occurred in the period from 2000 to 2018 were loaded into an open access data repository, specified and developed with the data recovery capacity required by the study. Cases were selected with both codes for the underlying cause of death (CB) and for the mentioned diagnosis (DM) considered valid according to the list of ICD-10 codes distributed on the SIM file transfer page and with a minimum of 2 and a maximum of 6 mentioned diagnoses (DMs). All sets of DMs from the data relating to the periods 2009-2013 and 2014-2018 were identified, representing them, for each DM, in six types of sets of DMs. Quantitative rules were then created, associating each set of DMs to the corresponding CBs (MD=>UCD Rules) from the respective proportional distributions. The Rules from the 2009-2013 period were then applied to each DO from deaths occurring in the period 2014 to 2018. And vice versa. Finally, the underlying causes of death predicted by the method were compared to those selected by SIM. Results: In the DO’s from 2009-2013, 5218 different underlying cause of death codes were selected and the following sets of mentioned diagnoses were identified: i) 635,995 distinct sets of DMs considering the lines of the death certificate and the order in which they were mentioned; ii) 596,544 disregarding the position, iii) 535,212 also disregarding the order of mentions; iv) 336,119 combinations of DM pairs; vi) 270,698 distinct pairs when disregarding the order of mentions; and vi) 7,430 different diagnosis codes. Combined with the respective CBs, there were 4,315,054 different sets of DMs x CBs, with 637,196, 611,208, 558,017, 1,196,629, 1,081,936 and 230,068 corresponding to these six types of DM sets. When applying the MD=>UCD Rules from the period 2009-2013 to the data from 2014-2018, underlying causes were assigned to 99.99% of the cases with 89.99% accuracy, at ICD-10 Subcategory level. When applying the MD=>UCD Rules from the period 2014-2018 to the 2009-2013 data, underlying causes of death were assigned to 99.99% of the cases with 90.36% accuracy, also at the ICD-10 Subcategory level. Conclusion: A Data Repository of Death Certificates registered in Brazil over the years 2000 to 2018 was made available, in open access and with data retrieval capacity also suitable for studies of multiple causes of death. The study concluded that the underlying cause of death of an MD is strongly associated not only with the information contained in the structured representation of the mentioned diagnoses, but also with other representations of them, such as sequences and pairs of MD, even disregarding the position and/or the order in which they were declared in the DO: underlying causes of death were predicted with a considerable degree of accuracy, simply through the proportional distributions of the sequences, pairs and individual mentioned diagnoses and their corresponding selected underlying causes of death. The “predicted” underlying causes of death may serve as a guide for manual verification of those manually selected and may be useful in revealing cases that require special attention – including in the decision tables of automated cause of death selection systems.
Introduction: Mortality statistics are produced in most countries, according to the rules defined by the WHO in successive revisions of the International Statistical Classification of Diseases and Related Health Problems, currently in its 10th Revision (ICD-10). Their main axis is the so-called Underlying Cause of death, selected in accordance with the International Rules for Selection of Underlying Causes, whose interpretation requires knowledge of causal relationships between diseases and considerable technical capacity to apply them correctly and systematically. Objective: The study proposed to identify the sets of diagnoses reported in data from Death Certificates processed by the Brazilian Mortality Information System (SIM), associate them with the corresponding selected underlying causes and evaluate the degree of accuracy of a statistical prediction method of these. Methods: Data from SIM Death Certificates (DO) with data from the State of São Paulo occurred in the period from 2000 to 2018 were loaded into an open access data repository, specified and developed with the data recovery capacity required by the study. Cases were selected with both codes for the underlying cause of death (CB) and for the mentioned diagnosis (DM) considered valid according to the list of ICD-10 codes distributed on the SIM file transfer page and with a minimum of 2 and a maximum of 6 mentioned diagnoses (DMs). All sets of DMs from the data relating to the periods 2009-2013 and 2014-2018 were identified, representing them, for each DM, in six types of sets of DMs. Quantitative rules were then created, associating each set of DMs to the corresponding CBs (MD=>UCD Rules) from the respective proportional distributions. The Rules from the 2009-2013 period were then applied to each DO from deaths occurring in the period 2014 to 2018. And vice versa. Finally, the underlying causes of death predicted by the method were compared to those selected by SIM. Results: In the DO’s from 2009-2013, 5218 different underlying cause of death codes were selected and the following sets of mentioned diagnoses were identified: i) 635,995 distinct sets of DMs considering the lines of the death certificate and the order in which they were mentioned; ii) 596,544 disregarding the position, iii) 535,212 also disregarding the order of mentions; iv) 336,119 combinations of DM pairs; vi) 270,698 distinct pairs when disregarding the order of mentions; and vi) 7,430 different diagnosis codes. Combined with the respective CBs, there were 4,315,054 different sets of DMs x CBs, with 637,196, 611,208, 558,017, 1,196,629, 1,081,936 and 230,068 corresponding to these six types of DM sets. When applying the MD=>UCD Rules from the period 2009-2013 to the data from 2014-2018, underlying causes were assigned to 99.99% of the cases with 89.99% accuracy, at ICD-10 Subcategory level. When applying the MD=>UCD Rules from the period 2014-2018 to the 2009-2013 data, underlying causes of death were assigned to 99.99% of the cases with 90.36% accuracy, also at the ICD-10 Subcategory level. Conclusion: A Data Repository of Death Certificates registered in Brazil over the years 2000 to 2018 was made available, in open access and with data retrieval capacity also suitable for studies of multiple causes of death. The study concluded that the underlying cause of death of an MD is strongly associated not only with the information contained in the structured representation of the mentioned diagnoses, but also with other representations of them, such as sequences and pairs of MD, even disregarding the position and/or the order in which they were declared in the DO: underlying causes of death were predicted with a considerable degree of accuracy, simply through the proportional distributions of the sequences, pairs and individual mentioned diagnoses and their corresponding selected underlying causes of death. The “predicted” underlying causes of death may serve as a guide for manual verification of those manually selected and may be useful in revealing cases that require special attention – including in the decision tables of automated cause of death selection systems.
Descrição
Citação
TARDELLI, Adalberto Otranto. Um método de seleção de causas básicas de óbito por meio das distribuições de diagnósticos mencionados. 2024. 100 f. Tese (Doutorado em Gestão e Informática em Saúde) – Escola Paulista de Medicina, Universidade Federal de São Paulo (UNIFESP). São Paulo, 2024.